A private AI platform
for your company
Hardware, models, assistants, integrations and operation from one team. Everything runs inside your perimeter, so contracts, code and customer data do not go to outside AI services.
Request a proposalYour data stays with you
Documents and prompts are processed inside your perimeter, with GDPR and the EU AI Act in mind.
No per-request bills
Your own or dedicated GPUs: costs you can plan, however much your teams use them.
Independent of providers
Open models on your infrastructure: no dependence on a vendor’s prices, terms or availability.
What the platform consists of
Select a layer to see what is included.
Assistants
What your staff work with every day.
- Search and answers over internal documents, with source and page
- An assistant for each department
- Agents that automate routine operations end to end
What your staff see
What notice period applies in the framework agreement with Müller GmbH?
Three months to the end of a quarter, in writing (§ 14.2). Without notice the agreement renews for another 12 months (§ 14.3).
- Framework agreement Müller 2024.pdf, p. 9
- Amendment 02.pdf, p. 2
Illustrative example: the companies, documents and answers are invented.
Where a request goes
Outside AI service
The prompt and any attached documents leave your company and are processed by a third party.
Your platform
The request is answered inside your perimeter. Nothing is sent outside.
Where it runs
We install and run the same platform in all three.
| Your hardware | Rented in the EU | Both | |
|---|---|---|---|
| Servers belong to | You | The data-centre operator; they are reserved for you | Part yours, part rented |
| They stand in | Your premises | An EU data centre | Both places |
| Costs | Purchase, then running costs | Rent, no purchase | A mix of both |
| Per-request charges | None | None | None |
| Fits when | Data rules are strict and the load is steady | You want to start without buying hardware, or the load may change | Part of the load stays with you, part runs on rented capacity |
How much hardware you need
| Model | Llama 3.3 70B |
|---|---|
| Simultaneous requests at peak | 10 |
| GPU memory needed | ≈ 92 GB |
| GPUs, rough estimate | 2 × H100 80 GB or equivalent |
| Servers (up to 8 GPUs each) | 1 |
An estimate for orientation, assuming H100-class GPUs and typical use. We size it precisely in the assessment, with your data and your peaks.
How a project runs
Assessment and architecture
Use cases, volumes, security and GDPR requirements. We choose the models and size the GPUs.
You getAn architecture and a sizing
Deployment
The GPU cluster, inference for open models, one API for your systems, role-based access and single sign-on.
You getA running cluster with API and sign-in
Your knowledge
Search and answers over internal documents, assistants for each department, connections to CRM, email and knowledge bases.
You getAssistants connected to your documents
Fine-tuning and agents
Models adapted to your terminology and processes; routine operations automated end to end.
You getModels that speak your language
Operation
Monitoring, model updates, audit logs, quality and cost control.
You getA platform that keeps running
Request a proposal
Tell us your use cases, how many people will use the platform and what data it will touch.